Networking equipment maker #Unifi is late to the #IPv6 game.

The first thing I come across is, that they build #NAT66 support which is neither standardized nor needed.

Why?

@tschaefer das Problem an #IPv6 ist dass es annimmt dass Leute statische Adressen haben und diese sich nicht nur nicht ändern sondern darüber erreichbar sind.

  • Schonmal #MultiWAN-Konnektivität stabil failovern gesehen? Bei #IPv4 werden schlimmstenfalls Verbindungen beendet und neu geöffnet.

  • Bei IPv6 ist dann erstmal alles kaputt - auch #LAN, wenn mensch kein #NAT66 mit #ULA macht!

@louis Once Provider-Independent #IPv6 isn't #paywalled with a @ripencc membership, #NAT66+#ULA works as well across #WAN links as #IPv4+#NAT and all vendors of Software and devices support IPv6 with Privacy Extensions enabled per default!

@0xIO32 ask @mook that.

As for #IPv4: Until #IPv6 is universally available and consumers get provider-independent address spaces or #NAT66 with #ULA|s just work it'll stick around…

I'm going to do the horrible thing and use #NAT66 since I only get one ipv6 /64 subnet per vps from my VPS provider, and I need two... #ipv6

@litchralee_v6

with internal #NAT66 DNAT and SNAT in an unwholesome combo.

All network engineers, in sympathy:
"I understand. I weep, but I understand."

After toiling for a few hours, I've managed to achieve a working #Palworld server on my #IPv6only #k8s cluster. The server is responsive on both modern and legacy IP, which I think might be fairly unique, since the game server doesn't listen on #IPv6 and I suspect the game client doesn't connect on IPv6 either.

This works using a Tundra-NAT64 CLAT sidecar, doing #464XLAT, with internal #NAT66 DNAT and SNAT in an unwholesome combo.

Occasionally, it pays to have a strong networking background.

In a cluster like #Kubernetes, would a Layer 3 load-balancer operating on #IPv6 traffic be considered a #NAT (specifically #NAT66)? There certainly is address translation happening but not necessarily port mapping. Would this be better described as round-robin routing, to fungible destinations?

I'm poised to think that it's not a NAT, because the goal is for the cluster to present a single face to the network, rather than to provide the cluster's nodes with IP connectivity.

and that advantage goes out the window with stateless one-to-one #NPTv6 so what's the advantage over stateful one-to-many #NAT66... considering YOU NEED A STATEFUL FIREWALL ANYWAY...